1 2 3 4 From assistance towards restoration with an implanted brain-computer interface based on epidural electrocorticography: A single case study or P 1 roo f Restorative Neurology and Neuroscience xx (20xx) x–xx DOI 10.3233/RNN-140387 IOS Press 7 a Division of Functional and Restorative Neurosurgery & Division of Translational Neurosurgery, Department 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 dA 13 15 cte 12 14 Abstract. Purpose: Today’s implanted brain-computer interfaces make direct contact with the brain or even penetrate the tissue, bearing additional risks with regard to safety and stability. What is more,these approaches aim to control prosthetic devices as assistive tools and do not yet strive to become rehabilitative tools for restoring lost motor function. Methods: We introduced a less invasive, implantable interface by applying epidural electrocorticography in a chronic stroke survivor with a persistent motor deficit. He was trained to modulate his natural motor-related oscillatory brain activity by receiving online feedback. Results: Epidural recordings of field potentials in the beta-frequency band projecting onto the anatomical hand knob proved most successful in discriminating between the attempt to move the paralyzed hand and to rest. These spectral features allowed for fast and reliable control of the feedback device in an online closed-loop paradigm. Only seven training sessions were required to significantly improve maximum wrist extension. Conclusions: For patients suffering from severe motor deficits, epidural implants may decode and train the brain activity generated during attempts to move with high spatial resolution, thus facilitating specific and high-intensity practice even in the absence of motor control. This would thus transform them from pure assistive devices to restorative tools in the context of reinforcement learning and neurorehabilitation. rre 10 11 of Neurosurgery, Eberhard Karls University Tuebingen, Tuebingen, Germany b Neuroprosthetics Research Group, Werner Reichardt Centre for Integrative, Neuroscience, Eberhard Karls University Tuebingen, Tuebingen, Germany c Department of Computer Engineering, Wilhelm-Schickard Institute for Computer, Science, Eberhard Karls University Tuebingen, Tuebingen, Germany d Department of Computer Engineering, University of Leipzig, Leipzig, Germany e Institute for Medical Psychology and Behavioural Neurobiology, Eberhard Karls, University Tuebingen, Tuebingen, Germany co 9 Un 8 uth 6 Alireza Gharabaghia,b,∗ , Georgios Narosa,b , Armin Walterc , Florian Grimma,b , Marc Schuermeyera,b , Alexander Rothc , Martin Bogdanc,d , Wolfgang Rosenstielc and Niels Birbaumeree 5 Keywords: Electrocorticography, neuroprosthetics, epidural implant, brain-computer interface, brain-machine interface, neurorehabilitation, stroke ∗ Corresponding author: Dr. Alireza Gharabaghi, Division of Functional and Restorative Neurosurgery, & Division of Translational Neurosurgery, Department of Neurosurgery Eberhard Karls University Otfried-Mueller-Str. 45, 72076 Tuebingen, Germany. Tel.: +49 7071 29 83550; Fax: +49 7071 29 25104; E-mail: alireza. gharabaghi@uni-tuebingen.de. 0922-6028/14/$27.50 © 2014 – IOS Press and the authors. All rights reserved 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 2.1. Patient roo f 39 2. Methods or P 38 The 68-year-old, male patient had suffered a hemorrhagic stroke of the left thalamus associated with pyramidal tract damage at the level of the posterior limb of the internal capsule (see Fig. 1) 34 months before implantation. This had led to intractable pain on the right side of his body, including arm, hand and face area, and a persistent right-sided severe hemiparesis associated with an increased muscle tone of the affected body side. In particular, he had also lost the functional use of his upper extremity (Medical Research Council motor scale <2). As a candidate for long-term cortical stimulation to reduce his chronic pain, the patient underwent implantation of an electrode array covering the left motor cortex to determine treatment response and optimal dA 37 cte 36 rre 35 Despite intensive rehabilitation training following evidence-based guidelines (Kwakkel et al., 2003), functional restoration in patients with severe and persistent motor deficits, for example after suffering a stroke, is very limited (Buma et al., 2013). Investigational studies are therefore currently exploring novel therapeutic strategies based on brain stimulation (Fusco et al., 2014; Kwon et al., 2013) or neurofeedback (Buch et al., 2008; Buch et al., 2012). In this vein, electroencephalography (EEG)-based neurofeedback training with brain-computer interfaces (BCI) has already been shown to successfully prime behaviorally oriented physiotherapy for motor rehabilitation in chronic stroke patients (Ramos-Murguialday et al., 2013). Such non-invasive BCI approaches are characterized by low spatial resolution, with each EEG contact sampling of the order of 105 –108 neurons, a low signalto-noise- ratio due to signal attenuation caused by the skull, possible contaminations by muscle artifacts and external electrical activity, and a relatively long training time to reach real-time control of devices (Leuthardt et al., 2009). By contrast, electrocorticographic (ECoG) BCI approaches may overcome these limitations thanks to their proximity to the neural signal source. Nonetheless, up until recently, ECoG devices in stroke patients have been implemented as assistive devices for prosthetic control (Yanagisawa et al., 2011) rather than for functional restoration in the context of neurorehabilitation. Moreover, current implantable BCIs in patients with motor deficits remain in direct contact with the brain via subdural grids (Yanagisawa et al., 2012; Wang et al., 2013) or even penetrate the tissue with intracortical microelectrodes (Hochberg et al., 2012; Collinger et al., 2013) bearing additional risks with regard to safety and stability in long-term application. Implantable but less invasive approaches might therefore enable us to devise novel rehabilitation tools for patients with severe motor deficits. Such tools would have to utilize spared neural networks to address the neurobiology of motor learning, e.g. by volitional control of cortical signals and facilitating practice with feedback and reward (Birbaumer and Cohen, 2007; Dobkin, 2007). We therefore implemented a neurofeedback rehabilitation set-up on the basis of an epidural implant that provides real-time feedback of motor-related electrocorticographic brain activity to operate a virtual grasping movement according to the principles of motor learning (Hebb, 1949). co 34 1. Introduction Un 33 A. Gharabaghi et al. / Restorative brain-computer interface uth 2 Fig. 1. T1-weighted axial MR image at the level of the lesion, indicating that the hemorrhagic stroke of the left thalamus also affected the posterior limb of the internal capsule. 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 109 110 111 112 113 114 115 116 117 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 During the training period, changes in the maximum amplitude of movement extension of the upper extremity were quantified by measuring the status with an ultrasonic motion detection system (Zebris Medical GmbH, Isny, Germany). The maximum movement extension was measured for each trial, and mean amplitude (+SD) for each session was calculated in five measurement sessions: one session 5 days before implantation (session 1), and four sessions on days 3 (sessions 2 and 3), 12 (session 4) and 24 (sessions 5) after implantation. On day 3 after implantation, one measurement session was performed before (session 2) and one measurement session (session 3) after the training session, respectively. The measurement sessions 1, 2, 3, 4, and 5 consisted of 14, 25, 15, 30, and 30 trials, respectively, and were independent of the BCI neurofeedback sessions. The motion detection system was based on ultrasound triangulation. Markers small enough to be attached to the hand enabled us to measure finger and wrist trajectories. During each measurement session, the patient was seated in a standardized fashion with a 45◦ angle between the upper arm and torso and a 90◦ angle in the elbow joint. For measurements of the wrist and fingers, the hand was positioned on a custom-built armrest. During measurements of the patient’s hand function, 20 markers were attached to the knuckles and fingertips of his hand as well as to his wrist and forearm. The patient was visually and acoustically cued to maximally extend and relax his hand in each run. On the basis of the position of the Fig. 2. Zoom out showing the implanted ECoG grid in relation to the reconstructed cortical anatomy of the patient. The grid covers the primary motor (M1) somatosensory (S1) and premotor cortex (PM). In addition, the orientation of the grid is projected onto a standard 3D brain reconstruction. The two contacts (green) used for the feedback training are projected onto the anatomical hand knob. markers, the amplitudes of different movement angles were calculated to indicate the range of motion during each task. cte 108 rre 107 co 106 Un 105 or P 2.2. Measurement of movement extension 104 uth 119 103 dA 118 stimulation sites for maximum pain reduction. Indication for implantation, array location and duration of implantation were determined solely by clinical criteria. The patient also gave informed consent to a study exploring a neurofeedback approach for motor restoration, which was conducted in accordance with the Declaration of Helsinki and the guidelines of the local ethics committee. Preoperative evaluation with electroencephalographic recordings and neurofeedback ensured that the patient was capable of controlling his oscillatory brain activity and that he was familiar with the experimental set-up. After implantation of the electrode array, the patient underwent seven training sessions with BCI neurofeedback on days 1, 2, 3, 4, 7, 8 and 9 after surgery. He performed 100–170 feedback trials per training session, resulting in a total of ∼900 feedback trials. 102 3 roo f A. Gharabaghi et al. / Restorative brain-computer interface 2.3. Implanted neural interface The epidurally implanted 8 × 12 electrode array consisted of platinum contacts with 4 mm contact diameter (2.3 mm exposed), and 5 mm center-to-center distance (Ad-Tech Medical Instrument Corp., Rancine, Wisconsin, USA), covering a broad area of the left motor cortex (arm, hand and face area) as well as parts of the premotor and somatosensory cortex (see Fig. 2). Following a two-week evaluation period, the array was removed and replaced by four permanent electrode leads for chronic application (Resume II, Medtronic, Minneapolis, USA) with four platinum iridium electrode contacts each (4 mm diameter, 10 mm center-to-center distance) at the sites where stimulation provided optimal pain control. This pain treatment is not part of the present report. 2.4. Neurofeedback set-up and training The integrated system (see Fig. 3) consisted of an internal component, the implanted neural interface for electrocorticographic (ECoG) recordings, externalized with percutaneous extensions connecting to the exter- 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 A. Gharabaghi et al. / Restorative brain-computer interface 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 cte 180 Prior to the neurofeedback sessions, we performed two screening sessions with approximately 60 trials to select channels and features for the BCI training, estimating the spectral power for all electrode channels up to 200 Hz, computing r2 scores and assessing their significance with a permutation test. The two channels with the highest r2 scores for discrimination between movement and rest for all channels located close to the sensorimotor hand representation were selected for the feedback sessions and were kept constant throughout the training period. During the training sessions, the spectral power of the two channels between 13 and 19 Hz was computed every 40 ms from a data buffer of length 500 ms. This entailed using an autoregressive model (Burg algorithm) with a model of the order of 16 (McFarland and Wolpaw, 2008). These power values were taken as input for an adaptive linear classifier, resulting in 25 classifier outputs per second. Five consecutive classifier outputs (200 ms) of the same feature were necessary to switch from one feedback condition to the other (e.g. to initiate virtual grasp movement). rre 179 co 178 Un 177 To assess the patient’s ability to modulate his brain activity contingent on the feedback task, i.e. his performance in gaining control of the BCI neurofeedback system, we calculated the average video movement time divided by the total feedback duration phase. We also measured a baseline condition to monitor natural perturbations of brain activity. Although not related to the task, these could have distorted the online BCI performance during the feedback task. This baseline condition consisted of several ECoG recordings while the patient was at rest with his eyes open throughout the 2-week study period. All in all, ∼30 minutes of this spontaneous baseline ECoG activity were recorded for offline analysis, segmented into trials of the same structure and processed in the same manner as the feedback sessions. For statistical analysis, the distribution of performance values per run in each feedback session was compared with the distribution of performance values for the baseline data (Wilcoxon rank-sum test). roo f 2.5. Brain-computer-interface 176 2.6. Performance evaluation or P 198 175 dA 197 nal components. These external components consisted of a recording and processing unit and of a feedback unit. Recording of ECoG signals was performed with a monopolar amplifier (Brain Products, Munich, Germany) with a high-pass filter at 0.15 Hz and a sampling rate of 1000 Hz. Online processing of brain signals was performed with a BCI 2000 framework (Schalk et al., 2004) extended with custom-built features to control a video player via UDP. The patient received a visual movement cue with an open hand for 4 seconds followed by a 4 second relaxation period and was instructed to parallel this cycle by imagining finger and wrist extension. Feedback was provided contingent on movement intention related oscillatory brain activity by a video player displaying a virtual grasp movement from the perspective of the patient when observing his hand (Minimal Video Player, Phonon API). There was a short break after every ten trials (i.e. after each run). The patient performed 100–170 feedback trials per training session. On days 1, 2, 3, 4, 7, 8 and 9 after implantation, one training session was performed, resulting in a total of seven training sessions with BCI neurofeedback and some 900 feedback trials in all. 174 uth 4 3. Results The epidural implant reliably detected electrocorticographic brain activity for providing real-time feedback of movement intention-related oscillations to control a virtual grasping movement. The brain signals, detected by the same two adjacent contacts with a center-to-center distance of only 5 mm projecting onto the anatomical hand knob of the affected hemisphere, were sufficient for consistent neurofeedback throughout the whole training period (see Fig. 2). Epidural recordings of field potentials revealed the beta-frequency band superior to alpha- and gamma-frequency bands for discrimination between movement intention and rest condition in this patient (see Fig. 4). The frequency band between 13–19 Hz was therefore applied in all training sessions for feedback. In each of the seven training sessions, the patient was able to modulate his brain activity contingent on the feedback task throughout the whole session. He thus acquired significant control of the BCI neurofeedback system. In fact, he controlled the virtual hand movement for approximately 60% of the duration of feedback in each session. His performance in controlling the feed-back device in an online closed-loop paradigm was therefore constant and was significantly 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 5 dA uth or P roo f A. Gharabaghi et al. / Restorative brain-computer interface Un co rre cte Fig. 3. Experimental set-up for reinforcement training with an implanted brain-computer interface including electrocorticography (ECoG) recording, signal amplification (Amp), software for online analysis and triggering (BCI2000) of visual feedback within a closed-loop framework. Fig. 4. Power/frequency plot for attempted hand movement (dotted lines) vs. rest (solid lines) for the two ECoG channels (black and gray) used in the feedback experiments (see also Fig. 2). We observed the anticipated desynchronization in the beta-band and the synchronization in the gamma-band. 266 267 268 269 higher than the baseline condition (29.3 ± 11%) throughout the training period (see Fig. 5). It took only seven training sessions, each lasting between 30 and 45 minutes, to significantly improve maximum wrist extension (which was quantified in separate measurement sessions) from 7.25◦ ± 2.09◦ (measurement session 1) before training, 6.0◦ ± 2.70◦ (measurement session 2) and 270 271 272 273 6 A. Gharabaghi et al. / Restorative brain-computer interface 4. Discussion Some essential questions with regard to the clinical viability of implantable BCI approaches remain unanswered. For instance, is a severely impaired patient with a central nerve system lesion able to gain onlinecontrol of a BCI even in the absence of volitional control of hand and arm movements? Are less invasive approaches for BCI control, such as epidural implants and few decoding channels, feasible in this group of patients? Could such devices be used as training tools for rehabilitation in the absence of physiotherapists and applied for repetitive exercises without having to periodically update the neural decoding? In the present study, we used an ECoG-based BCI set-up to address these questions. This implanted interface was a safe and feasible tool for neurofeedbackbased motor rehabilitation training of a chronic stroke patient with severe and persistent hemiparesis. This patient, who was otherwise unable to actively participate in conventional physiotherapy training of his affected upper extremity, could gain volitional control of the feedback device and thus engage himself in repetitive, high- intensity exercises of intending finger and wrist extension. He was able to participate in the exercises without the assistance of a physiotherapist, while monitoring online his own ability for volitional modulation of brain activity and receiving immediate reward for successful performance in this task by video feedback of the intended movement. Like earlier reports on able-bodied individuals (Leuthardt et al., 2006), we too succeeded in showing that epidurally recorded signals were appropriate for reliable BCI control in a motor-impaired patient, thereby further increasing the safety of such co rre cte dA Fig. 5. Percentage of average movement time of the virtual hand divided by the total feedback duration phase. The mean ± standard deviation of the performance measure per session is indicated by solid lines. The mean of the baseline data is indicated by a dotted line. An asterisk (*) marks sessions where the median of the performance measure differs significantly (p < 0.05) from the median of the baseline value (modified from Walter et al., 2012, presented at the 12th International Conference on BioInformatics and BioEngineering, Larnaca, Cyprus). uth or P roo f This improvement was specific for the trained muscle and was not observed in other paretic muscles which remained unchanged throughout the observation period, producing either fasciculations or no movements whatsoever. Feedback sessions and sessions for movement evaluation were performed independently of each other. The results therefore reveal a transfer and a real improvement of voluntary control. Since the last feedback training took place 9 days after implantation and the last two sessions for movement evaluation were performed 12 and 24 days after implantation, i.e. 3 and 15 days after the last feedback training, the movement gains reveal lasting motor learning. 274 275 276 277 278 279 280 281 282 Un Fig. 6. Comparison of measurements of maximum wrist extension taken by an ultrasonic motion detection system before (session 1), during (sessions 2 and 3) and after (sessions 4 and 5) the training sessions, respectively. 6.19◦ ± 1.45◦ (measurement session 3) during training to 21.58◦ ± 3.13◦ (measurement session 4) and 19.91◦ ± 3.1◦ (measurement session 5) after training (t-test between sessions 1,2,3 and 4,5: p < 0.0001, see Fig. 6). In summary, measurement session 3 was performed after three training sessions, and measurement 4 was performed after seven training sessions. This might well explain the performance change between measurements sessions 3 and 4. 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 A. Gharabaghi et al. / Restorative brain-computer interface 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 roo f 337 or P 336 uth 335 dA 334 types accordingly. Interestingly enough, in our case, the beta-frequency band was superior to the gammafrequency band for decoding. This might be related to the fact that, in the reported cases of spinal cord injury (Wang et al., 2013) and moderate brain lesion (Yanagisawa et al., 2011), the cortical physiology was less affected than in our case. Moreover, while these two approaches applied more complex decoding algorithms based on additional features such as a broader frequency band to control 3D cursor control (Wang et al., 2013) or different movement types (Yanagisawa et al., 2011), our less sophisticated approach, which was confined to a narrow frequency band (13–19 Hz), was sufficient for a robust discrimination between movement intention and rest condition for high-density training. This fact illustrates the major difference between the present approach and previous applications of ECoG BCI in motor impaired patients. Other approaches aimed at controlling external devices such as cursors or prosthetic arms as assistive tools, and did not yet strive to become neurorehabilitative tools for training and restoring lost motor function. Such restorative tools would have to address principles of motor learning such as facilitating practice with feedback and reward (Dobkin, 2007) while making use of the specific advantages of ECoG-based BCI technology. ECoG recordings allow for both temporally and spatially precise and robust online detection of movement imagery/intention in the absence of real movement (Miller et al., 2010). This paves the way for neurofeedback-based rehabilitation training for patients who lack volitional control of their limbs. The effectiveness of such approaches based on reinforcement learning would depend on factors such as need, effort, immediacy, contingency and desirable cost-benefit ratio and effort. The present set-up met the need of a severely impaired stroke survivor to actively engage in rehabilitation exercises in the absence of volitional motor control of his hand. The effort was deemed appropriate and motivating since the device enabled him to perform autonomous training without the assistance of a physiotherapist. Immediacy and contingency posed the greatest challenges to this tool. The BCI set-up enabled us to make an update of the control signal as often as every 40 ms. However, our algorithm did not provide visual feedback until 5 consecutive 40 ms epochs had been classified consistently, thus avoiding a noisy cte 333 rre 332 applications with intracranial recordings. Moreover, the same two adjacent electrode contacts were sufficient for BCI feedback throughout the training, indicating that smaller electrode grids with fewer contacts might suffice for future applications of this technique. These results tally with recent observations in epilepsy patients who had been subjected to a BCI visual speller with a single subdural contact (Zhang et al., 2013). Our findings therefore demonstrate the feasibility of less invasive and more straightforward approaches with only a small number electrode contacts in the epidural space for therapeutic applications of ECoG BCI. Furthermore, it was not necessary to adjust the parameterization between the recorded brain signals and the BCI control algorithms during the two-week training period. These findings tally well with a report on multiple-day ECoG BCI control with fixed parameters in an able-bodied patient with an implanted grid for seizure localization (Blakely et al., 2009). Similarly, we performed an initial screening and feature selection followed by a robust and stable-state control of the interface on the following training days. This is an important requirement for future real-life application of such tools for rehabilitation training, regardless of the necessity of periodical software adaptations by specialists at respective institutions. Despite the massive hemiparetic motor impairment of this patient due to stroke, channels projecting onto the anatomical hand knob of the affected hemisphere provided sufficient information about his movement intention to permit robust and stable BCI feedback training of the brain activity generated during the attempt to move. This indicates that the somatotopic organization of the affected hemisphere has been at least partially preserved. It also demonstrates that even patients suffering from severe central nerve system lesions are suitable recipients of implantable BCI devices that address restoration of the compromised brain hemisphere. ECoG studies in patients without motor deficits have shown movement-related spectral changes, i.e. decreases in the low-frequency band (8–32 Hz) with a larger cortical distribution and increases in the highfrequency band (76–100 Hz) with a more focused projection onto the sensorimotor cortex (Miller et al., 2007). The BCI ECoG cases in tetraplegia (Wang et al., 2013) and in moderate hemiparesis (Yanagisawa et al., 2011) have identified the gamma frequency band as the most informative band for decoding movement co 331 Un 330 7 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 roo f 436 This work was supported by grants from the German Research Council [DFG GH 94/2-1, DFG EC 307], Federal Ministry for Education and Research [BFNT 01GQ0761, BMBF 16SV3783, BMBF 03160064B, BMBF V4UKF014], and European Union [ERC 227632]. Parts of this work were presented by A. Walter et al. in November 2012 at the 12th International Conference on BioInformatics and BioEngineering, Larnaca, Cyprus. We thank Ramin Azodi Avval and Thiago Monteiro for their support with the figures. or P 435 Acknowledgments References 479 480 481 482 483 484 485 486 487 488 489 490 Birbaumer, N., & Cohen, L.G. (2007). Brain-computer interfaces: Communication and restoration of movement in paralysis. J Physiol, 579(Pt 3), 621-636. 491 Blakely, T., Miller, K.J., Zanos, S.P., Rao, R.P., & Ojemann, J.G. (2009). Robust, long-term control of an electrocorticographic brain-computer interface with fixed parameters. Neurosurg Focus, 27(1), E13. 494 Borton, D.A., Yin, M., Aceros, J., & Nurmikko, A. (2013). An implantable wireless neural interface for recording cortical circuit dynamics in moving primates. J Neural Eng, 10(2), 026010. 498 Buch, E., Weber, C., Cohen, L. G., Braun, C., Dimyan, M.A., Ard, T., Mellinger, J., Caria, A., Soekadar, S., Fourkas, A., & Birbaumer, N. (2008). Think to move: A neuromagnetic brain-computer interface (BCI) system for chronic stroke. Stroke, 39(3), 910917. 501 Buch, E.R., Shanechi, A.M., Fourkas, A.D., Weber, C., Birbaumer, N., & Cohen, L.G. (2012). Parietofrontal integrity determines neural modulation associated with grasping imagery after stroke. Brain, 135(2), 596-614. 506 Buma, F., Kwakkel, G., & Ramsey, N. (2013). Understanding upper limb recovery after stroke. Restorative Neurology and Neuroscience, 31(6), 707-722. 510 Collinger, J.L., Wodlinger, B., Downey, J.E., Wang, W., TylerKabara, E.C., Weber, D.J., McMorland, A.J., Velliste, M., Boninger, M.L., & Schwartz, A.B. (2013). High- performance neuroprosthetic control by an individual with tetraplegia. Lancet, 381(9866), 557-564. 513 Dobkin, B.H. (2007). Brain-computer interface technology as a tool to augment plasticity and outcomes for neurological rehabilitation. J Physiol, 579(Pt 3), 637-642. 518 Fusco, A., Iosa, M., Venturiero, V., De Angelis, D., Morone, G., Maglione, L., & Paolucci, S. (2014). After vs. priming effects of anodal transcranial direct current stimulation on upper extremity motor recovery in patients with subacute stroke. Restorative Neurology and Neuroscience, 1;32(2), 301-312. 521 Gharabaghi, A., Kraus, D., Leão, M.T., Spüler, M., Walter, A., Bogdan, M., Rosenstiel, W., Naros, G., & Ziemann, U. (2014). Coupling brain-machine interfaces with cortical stimulation for brain-state dependent stimulation: Enhancing motor cortex 526 dA 434 cte 433 rre 432 control signal or false positive feedback. This tradeoff between immediacy and contingency was sufficient both for robust BCI control and for the patient to comprehend that there is a direct link between his attempt to move and the virtual grasp movement. This is in agreement with earlier psychophysical studies in which participants who were repeatedly exposed to an artificially introduced 250 ms delay between voluntary actions and sensory consequences perceptually combined their voluntary actions with the sensory consequences and perceived that the delay was shortened by approximately 100 ms (Haggard et al., 2002). The most intriguing observation was that, after only a small number of training sessions of this reinforcement learning paradigm, an improvement in the function of the targeted muscle could be established. We propose that this is due to repetitive engagement of a visuomotor loop relevant to the specific movement. The contingent activation of the respective neural correlates strengthened the connection between brain activity generated during the attempt to move and movement execution up to a level resulting in voluntary movement. This connection might be further reinforced by applying brain stimulation during the BCI training (Gharabaghi et al., 2014). Due to the lack of cortical implants allowing for wireless BCI control, we devised an interface by connecting the intracranial implant to an external online processing framework for recording and BCI control. To this end, extension leads had to be externalized through the skin, thus limiting the possible duration of this set-up. Future clinical applications of this novel approach will require wireless devices capable of fast and reliable information transfer (Borton et al., 2013). This would enable us to perform long-term training periods of this paradigm in home-based environments, and thus provide the potential for more major motor recovery to improve the patients’ quality of life. In conclusion, epidural implants are a feasible tool for volitional control of cortical signals with high spatial resolution and, as such, facilitate specific and high-intensity practice with feedback and reward matching principles of motor learning. For patients with severe motor deficits, for example as a result of a stroke, such implanted brain- computer-interfaces may decode and train the brain activity generated during the attempt to move, even in the absence of motor control. This could transform them from pure assistive devices into restorative tools in the context of reinforcement learning and neurorehabilitation. co 431 Un 430 A. Gharabaghi et al. / Restorative brain-computer interface uth 8 492 493 495 496 497 499 500 502 503 504 505 507 508 509 511 512 514 515 516 517 519 520 522 523 524 525 527 528 529 A. Gharabaghi et al. / Restorative brain-computer interface 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 Ramos-Murguialday, A., Broetz, D., Rea, M., Laer, L., Yilmaz, O., Brasil, F.L., Liberati, G., Curado, M.R., Garcia-Cossio, E., Vyziotis, A., Cho, W., Agostini, M., Soares, E., Soekadar, S., Caria, A., Cohen, L.G., & Birbaumer, N. (2013). Brain-machine interface in chronic stroke rehabilitation: A controlled study. Ann Neurol, 74(1), 100-108. 569 Schalk, G., McFarland, D.J., Hinterberger, T., Birbaumer, N., & Wolpaw, J.R. (2004). BCI2000: A general-purpose brain-computer interface (BCI) system. IEEE Trans Biomed Eng, 51(6), 10341043. 575 Wang, W., Collinger, J.L., Degenhart, A.D., Tyler-Kabara, E.C., Schwartz, A.B., Moran, D.W., Weber, D.J., Wodlinger, B., Vinjamuri, R.K., Ashmore, R.C., Kelly, J.W., & Boninger, M.L. (2013). An electrocorticographic brain interface in an individual with tetraplegia. PLoS One, 8(2), e55344. 579 Yanagisawa, T., Hirata, M., Saitoh, Y., Goto, T., Kishima, H., Fukuma, R., Yokoi, H., Kamitani, Y., & Yoshimine, T. (2011). Real-time control of a prosthetic hand using human electrocorticography signals. J Neurosurg, 114(6), 1715-1722. 584 Yanagisawa, T., Hirata, M., Saitoh, Y., Kishima, H., Matsushita, K., Goto, T., Fukuma, R., Yokoi, H., Kamitani, Y., & Yoshimine, T. (2012). Electrocorticographic control of a prosthetic arm in paralyzed patients. Ann Neurol, 71(3), 353-361. 588 Zhang, D., Song, H., Xu, R., Zhou, W., Ling, Z., & Hong, B. (2013). Toward a minimally invasive brain-computer interface using a single subdural channel: A visual speller study. Neuroimage, 71, 30-41. 592 Hochberg, L.R., Bacher, D., Jarosiewicz, B., Masse, N.Y., Simeral, J.D., Vogel, J., Haddadin, S., Liu, J., Cash, S.S., van der Smagt, P., & Donoghue, J.P. (2012). Reach and grasp by people with tetraplegia using a neurally controlled robotic arm. Nature, 485(7398), 372-375. Kwakkel, G., Kollen, B.J., van der Grond, J., & Prevo, A.J. (2003). Probability of regaining dexterity in the flaccid upper limb: Impact of severity of paresis and time since onset in acute stroke. Stroke, 34(9), 2181-2186. Kwon, T.G., Kim, Y.H., Chang, W.H., Bang, O.Y., & Shin, Y.I. (2013). Effective method of combining rTMS and motor training in stroke patients. Restorative Neurology and Neuroscience, 1;32(2), 223-232. Leuthardt, E.C., Miller, K., Anderson, N.R., Schalk, G., Dowling, J., Miller, J., Moran, D.W., & Ojemann, J.G. (2007). Electrocorticographic frequency alteration mapping: A clinical technique for mapping the motor cortex. Neurosurgery, 60(4 Suppl 2), 260-270; discussion 270-271. Leuthardt, E.C., Miller, K.J., Schalk, G., Rao, R.P., & Ojemann, J.G. (2006). Electrocorticography-based brain computer interface the Seattle experience. IEEE Trans Neural Syst Rehabil Eng, 14(2), 194-198. roo f 537 Hebb, D.O. (1949). The Organization of Behavior: A Neuropsychological Theory. New York: Wiley. or P 536 565 uth 535 Miller, K.J., Schalk, G., Fetz, E.E., den Nijs, M., Ojemann, J.G., & Rao, R.P. (2010). Cortical activity during motor execution, motor imagery, and imagery-based online feedback. Proc Natl Acad Sci U S A, 107(9), 4430-4435. dA 534 Haggard, P., Clark, S., & Kalogeras, J. (2002). Voluntary action and conscious awareness. Nat Neurosci, 5(4), 382-385. cte 533 changes in cortical surface potentials during motor movement. J Neurosci, 27(9), 2424-2432. McFarland, D.J., & Wolpaw, J.R. (2008). Sensorimotor rhythmbased brain-computer interface (BCI): Model order selection for autoregressive spectral analysis. J Neural Eng, 5(2), 155162. Miller, K.J., Leuthardt, E.C., Schalk, G., Rao, R.P., Anderson, N.R., Moran, D.W., Miller, J.W., & Ojemann, J.G. (2007). Spectral rre 532 excitability for neurorehabilitation. Frontiers in Human Neuroscience, 8, 122. co 531 Un 530 9 563 564 566 567 568 570 571 572 573 574 576 577 578 580 581 582 583 585 586 587 589 590 591 593 594 595